Implement buffered reading [10x speed increase]

[suyashkumar]

After using pprof to run CPU profiles of the existing code (which I thought seemed slower than it should be), I noticed a lot of CPU time was going to syscalls. Given the file reading task that this code was performing this did not make sense (likey something I/O heavy going on here). After auditing the code I realized that the default readers in this codebase were unbuffered. Changing to buffered readers led to a 10x speed boost when reading a Native PixelData dicom from disk:

Before:

./build/dicomutil --extract-images-stream   13.86s user 9.69s system 106% cpu 22.204 total

After:

./build/dicomutil --extract-images-stream   2.91s user 0.38s system 159% cpu 2.061 total

pprof before:

(pprof) top5
Showing nodes accounting for 16.58s, 97.13% of 17.07s total
Dropped 69 nodes (cum <= 0.09s)
Showing top 5 nodes out of 43
      flat  flat%   sum%        cum   cum%
    15.46s 90.57% 90.57%     15.46s 90.57%  syscall.Syscall
     0.53s  3.10% 93.67%      0.58s  3.40%  runtime.mallocgc
     0.23s  1.35% 95.02%      0.23s  1.35%  runtime.greyobject
     0.20s  1.17% 96.19%      0.20s  1.17%  image.Rectangle.At
     0.16s  0.94% 97.13%      0.49s  2.87%  runtime.scanobject

pprof after change:

(pprof) top10
Showing nodes accounting for 2680ms, 92.10% of 2910ms total
Dropped 20 nodes (cum <= 14.55ms)
Showing top 10 nodes out of 82
      flat  flat%   sum%        cum   cum%
     930ms 31.96% 31.96%     1050ms 36.08%  runtime.mallocgc
     860ms 29.55% 61.51%     1510ms 51.89%  github.com/gradienthealth/dicom.readNativeFrames
     220ms  7.56% 69.07%      220ms  7.56%  image.(*Gray16).SetGray16
     210ms  7.22% 76.29%      210ms  7.22%  runtime.greyobject
     140ms  4.81% 81.10%      140ms  4.81%  syscall.Syscall
     110ms  3.78% 84.88%      360ms 12.37%  runtime.scanobject
      60ms  2.06% 86.94%      900ms 30.93%  main.generateNativeImage
      60ms  2.06% 89.00%       60ms  2.06%  runtime.memclrNoHeapPointers
      50ms  1.72% 90.72%       50ms  1.72%  runtime.memmove
      40ms  1.37% 92.10%       40ms  1.37%  runtime.heapBitsForObject

[suyashkumar]

dd